Thanks for the explanation. I was thinking in terms of 'mining ensembles' from scratch, whereas you were thinking in terms of combining existing systems. It makes more sense now

Machine Learning with algoTraderJo

But surely to construct the map you would need to backtest every possible ensemble within the parameter constraints, since we do not know in advance which ensembles lead to which sequence? On an abstract level, it's not the calculation of the ...

Machine Learning with algoTraderJo

I'm not sure that I understand exactly what you mean. Particularly, what do you mean by "So at the end we have to calculate the trading criterion (e.g. sharpe ratio) only for a reduced number of sequences."? Besides, when testing on 28.5 years of ...

Machine Learning with algoTraderJo

No I haven't done it yet either. I will try when I get the chance... might be a couple of days though

Machine Learning with algoTraderJo

With regards to my post above (#625), Daniel from Asirikuy pointed out that this method does not achieve the desired result since, as algoTraderJo mentioned, often the best ensembles are comprised of systems with inferior individual performance. ...

Machine Learning with algoTraderJo

I would be hesitant to use only a single random categorisation which then, as you say, necessitates repeating the process and averaging the result. I suppose we would need to investigate the added computational cost which this method entails (on ...

Machine Learning with algoTraderJo

Now suppose we have found 10 prospective systems (parameter sets) from the original search space which satisfy our performance objective(s) and each have a low DMB. We wish to create an ensemble model. Could we conduct a brute-force optimisation ...

Machine Learning with algoTraderJo

Yes this is what I thought. For arguments sake, suppose our model has 5 parameters each ranging in the search space from 1 to 100 in steps of 10. Also suppose that we wish to construct an ensemble of dimension 10. Then, as you say, we can restrict ...

Machine Learning with algoTraderJo